Assessment of Density Functional Methods for Geometry Optimization

May 15, 2018 - Molecular Structure and Design, Bristol-Myers Squibb Company, P.O. Box 5400, ... *E-mail: [email protected]. ... For these functional...
0 downloads 0 Views 9MB Size
Subscriber access provided by UNIV OF DURHAM

Quantum Electronic Structure

Assessment of Density Functional Methods for Geometry Optimization of Bimolecular van der Waals Complexes Dominic A Sirianni, Asem Alenaizan, Daniel L. Cheney, and C. David Sherrill J. Chem. Theory Comput., Just Accepted Manuscript • DOI: 10.1021/acs.jctc.8b00114 • Publication Date (Web): 15 May 2018 Downloaded from http://pubs.acs.org on May 16, 2018

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

Assessment of Density Functional Methods for Geometry Optimization of Bimolecular van der Waals Complexes Dominic A. Sirianni,† Asem Alenaizan,† Daniel L. Cheney,‡ and C. David Sherrill∗,¶ †Center for Computational Molecular Science and Technology, School of Chemistry and Biochemistry, Georgia Institute of Technology, Atlanta, GA 30332-0400 ‡Molecular Structure and Design, Bristol-Myers Squibb Company, P.O. Box 5400, Princeton, New Jersey 08543 ¶Center for Computational Molecular Science and Technology, School of Chemistry and Biochemistry, and School of Computational Science and Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0400 E-mail: [email protected]

Abstract We explore the suitability of three popular density functionals (B97-D3, B3LYPD3, M05-2X) for producing accurate equilibrium geometries of van der Waals (vdW) complexes with diverse binding motifs. For these functionals, optimizations using Dunning’s aug-cc-pVDZ basis set best combine accuracy and a reasonable computational expense. Each DFT/aug-cc-pVDZ combination produces optimized equilibrium geometries for 21 small vdW complexes of organic molecules (up to four non-hydrogen atoms total) that agree with high-level CCSD(T)/CBS reference geometries to within

1

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

±0.1 ˚ A for the averages of the center-of-mass displacement and the mean least rootmean-squared displacement. The DFT/aug-cc-pVDZ combinations are also able to reproduce the optimal center-of-mass displacements interpolated from CCSD(T)/CBS radial potential energy surfaces in both NBC7x and HBC6 test sets to within ±0.1 ˚ A. We therefore conclude that each of these denisty functional methods, together with the aug-cc-pVDZ basis set, are suitable for producing equilibrium geometries of generic non-bonded complexes.

1

Introduction

Structure-based computer-aided drug design (SB-CADD) has emerged as a valued approach in the development of novel pharmaceutical compounds. Optimization of binding affinity of a lead chemical series is an iterative process, often requiring a detailed understanding of existing host-guest interactions and the ability to successfully predict new ones. These assessments are typically performed with molecular mechanics forcefields in combination with structural models based on crystallographic structures of closely related molecules. Such methods, however, are not always sufficiently fine-grained to accurately describe or quantify the guest-host interactions of interest. Capable of augmenting the existing SB-CADD paradigm by providing the information necessary to allow for the rational refinement of the resulting drug candidates, ab initio quantum-chemical methods offer a first-principles description of the non-covalent interactions (NCI) which govern host-guest binding. In particular, energy decomposition analysis (EDA) schemes such as the absolutely localized molecular orbital EDA (ALMO-EDA) 1–4 and symmetry-adapted perturbation theory (SAPT) 5–8 approaches offer a physically meaningful breakdown of interaction energies into contributions from more fundamental components, such as electrostatics, induction, dispersion, and exchange-repulsion. Furthermore, the atomic 9 and functional-group 10 partitionings of SAPT (A-SAPT and F-SAPT, respectively) offer an additional layer of interaction energy decomposition into the specific interactions between pairs of atoms or functional groups on each inter2

ACS Paragon Plus Environment

Page 2 of 34

Page 3 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

acting species. Indeed, these methods have already provided insight into the relative stability of chlorinated vs. methylated factor Xa inhibitors 11 and the role of NCI on transition-state stabilization in organocatalyzed aldol addition. 12 Before performing any of these quantum-chemical computations within SB-CADD applications, a model system must first be constructed which mimics the NCI of interest, and a geometry of suitable quality obtained, as resolution of tenths of kcal/mol or less may be necessary to distinguish between relevant binding configurations (e.g., the sandwich and T-shaped configurations of the pyridine dimer differ in interaction energy by only 0.1 kcal mol−1 ! 13,14 ) or seemingly minor chemical modifications. While significant attention has been paid to the optimization of individual molecules, a general protocol that is capable of generating accurate geometries of supramolecular assemblies is conspicuously lacking in the literature. Even among benchmark sets of non-covalently bound complexes, significantly more consideration is given to the computation of interaction energies than to the geometry optimization of the complexes themselves. 15–23 Hence, despite the availability of high-quality interaction energies approaching the coupled-cluster through perturbative triples [CCSD(T)] 24 complete-basis set (CBS) limit, there are very limited data on high-quality geometries of van der Waals dimers that might be used to assess various approximate methods for geometry optimization. Shown previously to be capable of reproducing benchmark quality IEs to within sub–kcal mol−1 accuracy 20,25,26 while maintaining low execution time relative to post-Hartree–Fock electron correlation methods like second-order Møller–Plesset perturbation theory (MP2) or CCSD(T), density functional theory approaches that include a treatment of London dispersion forces seem like natural candidates for routine application to the geometry optimization of such complexes. Indeed, some of the best such approaches yield MAE of only a few tenths of one kcal mol−1 for the S22 test set. 25 We therefore explore here the suitability of three of the best of these methods (B97-D3, 27 B3LYP-D3, 28,29 and M05-2X 30 ), where -D3 denotes the third-generation dispersion correction of Grimme; 31 these functionals have exhibited MAD = 0.48, 0.79, and 0.36 kcal mol−1 , respectively, for non-counterpoise–corrected interaction

3

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

energies versus benchmarks for complexes in the S22 test set. 25 The performance of each functional is assessed first by comparing DFT-optimized geometries for the 21 minimum-energy complexes in the A24 test set of Hobza and co-workers, 32 for which CCSD(T)/CBS–quality geometries are available due to the small size of these complexes (A24 systems contain up to four non-hydrogen atoms). Fully-optimized CCSD(T)/CBS geometries of larger systems would be computationally difficult to obtain; however, here we also present CCSD(T)/CBS potential energies vs intermolecular separation for 13 systems with up to twelve non-hydrogen atoms. These one-dimensional potential energy curves allow us to assess density functionals for their ability to reproduce the optimal intermolecular separation in these larger complexes, in order to validate the conclusions drawn from the optimization of the A24 systems.

2

Computational Methods

Throughout this study, we employ three density functionals which have become routinely applied to NCI: 20,25,26 B97-D3 (generalized gradient approximation, GGA), 27 B3LYP-D3 (hybrid-GGA), 28,29 and M05-2X (hybrid-meta–GGA); 30 each of these is paired with the popular correlation-consistent basis sets of Dunning both with and without augmentation by diffuse functions [(aug-)cc-pVXZ, X = D, T; abbreviated throughout as aXZ and XZ, respectively]. The relative computational expense of each of these functionals increases with each successive rung, from B97 → B3LYP → M05-2X, with B97 exhibiting an order of magnitude smaller overall algorithmic scaling compared to both B3LYP and M05-2X when density fitting is employed [O(N 3 ) versus O(N 4 ), respectively, with N proportional to overall system size]. This increase in computational scaling results from an increase in the amount of physics recovered by each successive functional: GGAs, which depend only on the gradient of the density, incorporate only local correlation; hybrid-GGAs incorporate a percentage of Hartree–Fock exchange, recovering some nonlocal correlation; and hybrid-meta–GGAs incorporate exact exchange in addition to a functional dependance on both the gradient

4

ACS Paragon Plus Environment

Page 4 of 34

Page 5 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

and Laplacian of the local density, recovering both nonlocal correlation and a more correct description of the topological dependence of the energy on the overall electron density. For the interested reader, we have included in Section I C in the supplementary information a brief summary of timings for the construction of these gradients with the methods examined here. We have applied the -D3 dispersion correction of Grimme 31 to B97 and B3LYP, as correction for missing dispersion in these functionals has been shown to be necessary for a high-quality description of NCI. 25 We have chosen this pairwise dispersion treatment as opposed to a many-body, 33–37 exchange-dipole moment, 38–40 or non-local 41–44 approaches due to the availability of low-cost analytical gradients for the -D family of corrections, allowing for minimal additional expense when incorporated into the geometry optimization procedure. The -D3 correction should give correct long-range behavior for the London dispersion interactions, while also accounting for the local chemical environment around each atom. 31 M05-2X, on the other hand, can describe London dispersion interactions at short to intermediate distances (up to ∼5 ˚ A), 45 but fails to have correct long-range behavior. This deficiency should not be significant for the smaller molecular systems examined here, but can become a problem for large systems with many long-range contacts. 46,47 For a thorough discussion of the ladder of approximations within density functional theory, a recent review of dispersion corrections in DFT and other mean-field electronic structure methods, and the application of density functional theory to study noncovalent interactions, we refer the interested reader to Refs. 48, 49, and 25, respectively, and the references therein.

2.1

Optimized Geometries for A24 systems

Geometries were optimized for the 21 minimum-energy complexes in the A24 test set of Hobza and co-workers 32 (complexes 1–21, denoted A21; visualized in Fig. 1.a) using the dispersion corrected functionals described above. Optimizations with B3LYP-D3 and B97D3 used a development version of the open-source Psi4 electronic structure program, 50 and 5

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

optimizations with M05-2X used the Q-Chem 4 program package. 51 In both cases, full optimizations were performed, allowing for monomer relaxation. For the optimization of these systems, we did not attempt to correct for basis set superposition error (BSSE), as this would add computational expense and would also be more difficult to automate with standard geometry optimizers. Fortunately, our results indicate that BSSE correction is not necessary for reliable geometries. Convergence criteria used for optimizations in this work were a) the energy difference between successive optimization steps below 1 × 10−6 Eh , b) the maximum component of the gradient below 1.5 × 10−5 Eh /a0 , c) the root-mean-square of the elements of the gradient below 1.0 × 10−5 , d) the maximum atomic displacement between successive optimization steps below 6.0 × 10−4 a0 , and e) the root-mean-square of the atomic displacements between successive optimization steps below 4.0 × 10−4 . For optimizations performed using Psi4, each of these five criteria must be satisfied for convergence to be achieved; for those performed using Q-Chem, however, convergence was achieved when criterion (b) and either of (a) or (d) were satisfied. These thresholds were chosen such that the difference between optimized geometries were less on average than the differences between the average errors of the DFT methods chosen. As meta-GGAs have been shown to exhibit oscillations in intermolecular potential energy surfaces of dispersion-bound complexes, 52 we have adopted a dense integration grid (150 radial points, 434 spherical points) for all density functional computations. To reduce the computational expense of DFT computations incurred by employing dense integration grids, the density-fitting approximation was applied to the electron repulsion integrals for computations performed using Psi4. 53–60 To examine the effect of basis set on the quality of the optimized geometries, we have employed Dunning’s correlation consistent polarized valence basis sets, 61 both with (aug-cc-pVXZ; X = D, T) and without (cc-pVXZ; X = D, T) augmentation by diffuse functions. For the convenience of the reader, these basis sets will be abbreviated as aXZ and XZ, respectively. For this work, we take the originally published geometries for each A21 complex as bench-

6

ACS Paragon Plus Environment

Page 6 of 34

Page 7 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

marks. 62 These were optimized by minimizing the numerical gradient of the counterpoisecorrected (CP) CCSD(T) interaction energy for each complex, at the complete basis set (CBS) limit. These interaction energies were estimated using the popular focal-point composite approach, 63,64 whereby the CBS limit estimate of the total MP2 energy [computed using the two-point extrapolation scheme of Helgaker, 65 denoted as MP2/CBS(aXZ, a[X+1]Z)], is corrected for higher-order correlation effects by adding the difference between CCSD(T) CCSD(T)

and MP2 as computed in a smaller basis set (denoted δMP2

). In particular, these bench-

CCSD(T)

marks were computed at the MP2/CBS(aTZ, aQZ) + δMP2

/aDZ level; this treatment

will be denoted here as CCSD(T)/[aTQZ; δ:aDZ]. This approach has been widely applied to estimate the CCSD(T) complete basis set limit for interaction energies. 21,23,25,26,66–77 The quality of optimized supramolecular geometries computed using each model chemistry examined here is assessed according to the following metrics: (i) the center-of-mass displacement (∆COM) between the monomers comprising each complex, compared to the ∆COM in the corresponding benchmark geometry, and (ii) the least root mean square deviation (LRMSD) between the optimized geometry and the accompanying benchmark geometry. For the purposes of this work, we consider both ∆COM and LRMSD metrics with values less than 0.1 ˚ A to correspond to “satisfactory” optimizations. However, we expect that for in many applications, larger errors of LRMSD ≈ 0.15–0.2 ˚ A and ∆COM ≈ ± 0.1– 0.15 ˚ A could remain acceptable. A more detailed discussion regarding the assessment of optimization quality using the metrics listed above and the choice of optimization thresholds is presented in Sections I A & B of the Supporting Information.

7

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Ammonia·Water

1

Water2

2

–6.506

Methane·Water

8

3

–5.015

Formaldehyde2

9

–0.665

(HF)2

(HCN)2

–4.751 Water·Ethene

10

–4.565

4

Ammonia2

–4.502

11

–1.626

Methane·Ethene

Borane·Methane

Methane·Ethane

Methane·Ethane

15

16

17

18

–0.504

–1.493

–0.830

–0.609

Methane·HF

–3.142 Ethine2

12

6

–0.534

–1.094 Ethene·Ar

21

–0.406

–0.354

(a) A21 Formic Acid2

Formamide2

R

R 1 –19.9 kcal/mol

2

–16.6

–2.7

R

Formamidine2

Formic Acid·Formamide

2 Bz·H2S

S2 Pyridine2

Methane2

Bz·Methane

T3 Pyridine2

3

–0.5

–2.8

–2.9

5

–16.6

4 –18.6

Formamide·Formamidine

–1.5

4

14

–1.377

20

–0.767 Ethene2

Methane·Ar

–1.7 kcal/mol

1

7

–1.661

13

Methane2

19

Methane·Ammonia

Ammonia·Ethene

–1.527

(d)

T-shaped Bz2

Sandwich Bz2 R

5

Formaldehyde· Ethene

–2.564

Page 8 of 34

6

7

Formic Acid·Formamidine

–3.0

8

5

–16.9

(b) NBC7x

6

–26.1

(c) HBC6

Figure 1: Test sets of bimolecular complexes examined here. (a) A21: 21 bound complexes contained in the A24 test set of Hobza and co-workers, 62 (b) NBC7x: seven (recently extended 13 ) radial potential scans from the NBC10 test set 14 and (c) HBC6: 22,23 radial potential scans for six doubly hydrogen bonded complexes. Indicated by box coloring [(a)–(c)] or by dot color [(d)] are the noncovalent interaction type for each complex, reported previously: 26,78 red for electrostatic interactions, blue for dispersion interactions, and yellow/green for mixed electrostatic and dispersion interactions.The ternary diagram (d) further indicates the relative magnitude of the interaction energy components for these complexes, 22,79 by placing a colored dot according to the ratios of attractive dispersion/induction and attractive/repulsive electrostatic contributions to the total interaction energy. Proximity to each labeled vertex indicates an increasing fraction of the attraction (repulsion) arising from that component.

8

ACS Paragon Plus Environment

Page 9 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

2.2

Radial Potential Surface Scans of HBC6 and NBC10x systems

In order to assess the generality of the conclusions drawn for the geometry optimizations of the A21 test set, the ability of DFT to reproduce optimal intermolecular separation distances between monomers in complexes from the NBC7x 13,14 and HBC6 22,23 test sets (visualized in Fig. 1b & c, respectively) was examined. To do this, radial potential energy surface scans of the selected bimolecular complexes were constructed from both counterpoise-corrected 80 (CP) and uncorrected (unCP) interaction energies computed with each combination of density functional and basis set examined above, using a 0.1 ˚ A step size. In order to estimate the optimal intermolecular separation for each curve, a second-degree polynomial was fit to the three (R, IE) points straddling the well minima using the Numerical Python (NumPy) library, 81 which was subsequently used to interpolate the optimal center-of-mass displacement (Req ) for each complex. We use here the label Req to distinguish these interpolated intermolecular separation distances from the center-of-mass displacements (∆COM) reported for optimized A21 complexes, since (i) the separation coordinate used to construct curves in the HBC6 and NBC7x test sets does not necessarily coincide with the vector connecting monomer centers-of-mass, and (ii) to further differentiate these interpolated distances computed from interaction energy curves from the monomer center-of-mass displacements within fully optimized structures. Benchmark values for the Req corresponding to each curve were determined by applying this same curve-fitting procedure to the HBC6 revision A 23 and NBC10 revision B 13 reference curves, respectively, each constructed from energies computed at the CCSD(T)/CBS limit. Information on these interaction energy benchmarks and revisions for HBC6 and NBC7x can be found in Table S-1 of the Supporting Information. For the formic acid dimer (HBC6-1) with all DFT model chemistries, and for the formic acid–formamidine complex (HBC6-6, see Fig. 1) with all M05-2X model chemistries, radial curves do not exhibit clear potential wells within which Req could be interpolated; we have therefore removed these curves from 9

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

the statistical analysis visualized in Fig. 5 and discussed in Section 3.2. We have, however, included these curves in the SI (see Section 6).

3

Results and Discussion

We first assess each model chemistry for its ability to produce optimal geometries for A21 complexes, in Section 3.1, before examining the generality of these conclusions in Section 3.2 by computing the optimal center-of-mass displacements for radial potential energy curves of larger bimolecular complexes in the HBC6 and NBC7x test sets.

3.1

Optimization of A21 Systems

As can be seen in Fig. 2.a & b for the B3LYP-D3 and B97-D3 density functionals, respectively, the aDZ basis set yields the best results for LRMSD values (light blue box-and-whisker plots) for A21 complexes. Indeed, for every density functional paired with aDZ, the average value of LRMSD, µLRMSD, is ≤ 0.05 ˚ A and, for all A21 complexes except a single outlier (LRMSD = 0.15 ˚ A for the water–ethene complex with M05-2X/aDZ; see Fig. 2.c), LRMSD ≤ 0.1 ˚ A. The aTZ basis set also exhibits good performance, with inner quartile ranges (IQRs) of less than 0.03 ˚ A for each density funtional. Despite this good performance for the majority of A21 systems, four complexes optimized with B97-D3/aTZ exhibit LRMSD ≥ 0.1 ˚ A [methane–water (A24-8; LRMSD = 0.5 ˚ A), ammonia–ethene (A24-13; LRMSD = 0.1 ˚ A), Cs methane–ethane (A24-15; LRMSD = 0.2 ˚ A), and borane–methane (A24-16; LRMSD = 0.5 ˚ A)], as opposed to just a single complex (Cs methane–ethane, A24-15; LRMSD = 0.2 ˚ A) for B3LYP-D3/aTZ, and no such complexes for M05-2X/aTZ. B3LYP-D3/aTZ and M052X/aTZ have slightly smaller IQRs and µLRMSD values than for aDZ, but the improvement is quite small. For M05-2X, a more noticeable improvement is observed in the overall range of LRMSD values, which decreases from 0.14 ˚ A for aDZ to only 0.06 ˚ A for aTZ. For the ∆COM metric, visualized in Fig. 2 with light red box-and-whisker plots, signed

10

ACS Paragon Plus Environment

Page 10 of 34

Page 11 of 34

0.5 Å

0.2

0.7 Å

(a) B3LYP-D3

0.1 0.0 –0.1 –0.2

0.5 Å

0.5 Å

0.2

0.5 Å

(b) B97-D3

0.1 0.0 –0.1 –0.2

0.2

0.6 Å

(c) M05-2X 0.6 Å

A21 ∆COM Signed Error & LRMSD (Å)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

0.1 0.0 –0.1 –0.2

A21 ∆COM Signed Error A21 LRMSD

DZ

TZ

aDZ

aTZ

Figure 2: Box-and-whisker plots representing both ∆COM signed errors (boxes shaded pink) and LRMSD values (boxes shaded blue) for systems in the A21 test set, optimized using the (a) B3LYP-D3, (b) B97-D3, and (c) M05-2X density functionals together with the DZ, TZ, aDZ, and aTZ basis sets. Boxes encompass the first (Q1) through third (Q3) quartiles of each data set, with values corresponding to the median (Q2) and mean LRMSD and ∆COM signed error indicated as a solid black bar and black square, respectively. Whiskers encompass the full range of LRMSD values and ∆COM signed errors; maximum values are indicated when whiskers surpass the area shown. For reference, a dotted line indicates 0.0 ˚ A, and three levels of shading are provided: light grey encompassing ±0.1 ˚ A, medium grey ˚ ˚ encompassing ±0.05 A, and medium-dark grey 11 encompassing ±0.01 A. ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

errors (SE) for individual model geometries and mean signed errors (MSEs) over all A21 complexes are not so clearly superior for the aDZ basis as was observed for the LRMSD metric, and ∆COM IQRs seem to be largely comparable between basis sets for each density functional. In fact, for both B3LYP-D3 and M05-2X, while the overall ranges of ∆COM signed errors are slightly smaller when using the aDZ basis set, the MSEs for these functionals are smallest with the aTZ basis set. For B97-D3, each of the aDZ, TZ, and aTZ basis sets yield error ranges which are nearly identical, with maximum and minimum signed errors lying slightly outside the target range of ±0.1 ˚ A. The mean signed errors for these combinations benefit from this nearly symmetric distribution; both B97-D3/aDZ and B97-D3/TZ exhibit MSE ≤ ±0.01 ˚ A, and B97-D3/aTZ is not much worse, with MSE = 0.03 ˚ A. Generally, however, the relative quality of model geometries with respect to the ∆COM metric is again not significantly improved when moving from aDZ to aTZ basis sets. IQRs improve slightly for each functional; however, the overall range of ∆COM values increases for both B3LYP-D3 and M05-2X functionals. Regardless of this increase in the total range for these model chemistries, the highly symmetric distribution of ∆COM values about 0.0 ˚ A signed error yields very small MSEs, with MSE = -0.01, 0.00 ˚ A for B3LYP-D3/aTZ and M05-2X/aTZ; the double-ζ counterparts are not much worse, however, with MSE = -0.05, -0.04 for B3LYP-D3/aDZ and M05-2X/aDZ. B97-D3, on the other hand, exhibits the opposite trend when moving from aDZ to aTZ: while the total range of ∆COM values improves from 0.32 ˚ A to 0.26 ˚ A , MSE increases slightly, from MSE = 0.00 ˚ A to MSE = 0.03 ˚ A. The similar or only marginally improved performance of optimizations utilizing aTZ over aDZ for both LRMSD and ∆COM metrics, together with the increased computational cost associated with the increase in ζ-level, implies that the smaller aDZ basis set is generally preferable for optimizations of nonbonded complexes similar to those within the A21 test set. To more closely examine the performance of each density functional for optimizing A21 complexes, we next consider the quality of individual equilibrium geometries optimized using

12

ACS Paragon Plus Environment

Page 12 of 34

(a) LRMSD 0.12 0.08 0.04 0.00 0.15

∆COM Signed Error (Å)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

LRMSD of Optimized Geometry (Å)

Page 13 of 34

(b) ∆COM

HB Subset Members MX Subset Members DD Subset Members A21 Mean LRMSD

0.05

–0.05

–0.15 B3LYP-D3 aug-cc-pVDZ

B97-D3 aug-cc-pVDZ

M05-2X aug-cc-pVDZ

Figure 3: Values corresponding to individual A21 complexes and box-and-whisker plots detailing test-set–wide distributions for (a) least root mean square displacements (LRMSD) and (b) signed errors in center-of-mass distance (∆COM), computed with each density functional using the aug-cc-pVDZ basis set. Values for individual A21 systems, shown with empty circle markers, are gouped and colored according to interaction motif: 26,78 red for electrostatically bound complexes (HB subset), blue for dispersion bound complexes (DD subset), and green for mixed interaction complexes (MX subset). For box-and-whisker plots of each model chemistry, boxes encompass the first (Q1) through third (Q3) quartiles of each data set, with values corresponding to the median (Q2) and mean LRMSD and ∆COM signed error indicated as a solid green bar and black square, respectively. Whiskers encompass the full range of LRMSD values and ∆COM signed errors; maximum values are indicated when whiskers surpass the area shown. For reference, a dotted line indicates 0.0 ˚ A, and three ˚ levels of shading are provided: light grey encompassing ±0.1 A, medium grey encompassing ±0.05 ˚ A, and medium-dark grey encompassing ±0.01 ˚ A.

13

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

the recommended basis set, aDZ. Visualized in Fig. 3 are values corresponding to each A21 complex, and box-and-whisker plots describing these values’ distributions for (a) LRMSD and (b) ∆COM metrics. Within each density functional, electrostatically bound complexes (HB subset, red circles) exhibit LRMSD values and ∆COM signed errors that are more clustered than for the other two A21 subsets [dispersion-dominated (DD) subset (blue circles) and mixed-interaction (MX) systems (green circles)]. Among these density functionals, M05-2X/aDZ generates model geometries that are notably superior for HB and DD complexes, with respect to both LRMSD and ∆COM metrics; this model chemistry is also slightly superior to B3LYP-D3/aDZ and B97-D3/aDZ for MX systems, with the lone exception being the water–ethene complex, exhibiting LRMSD = 0.15 ˚ A. For B3LYP-D3/aDZ, Req is underestimated in nearly all A21 model geometries (negative signed error for ∆COM), while M05-2X/aDZ generally underestimates Req for A21 complexes. B97-D3/aDZ model geometries, on the other hand, exhibit different behavior for the ∆COM metric depending on the interaction type of the complex; Req is typically overestimated in HB systems, underestimated in DD systems, and no trend is observed for the MX subset. Based on the performance of these model chemistries for producing optimal geometries of A21 test set, a total ordering which ranks the performance of the best such model chemistries can be constructed. Here, we consider first the µLRMSD and ∆COM MSE statistics for each set of values, then the sample inner quartile and total ranges; these considerations produce the following ordering: M05-2X/aTZ ∼ B3LYP-D3/aTZ % M05-2X/aDZ % B3LYP-D3/aDZ  B97-D3/aDZ where “∼” indicates roughly equivalent performance of sample means and IQR, “%” indicates superior performance with respect to sample mean, but roughly similar performance in IQR, and “” indicates superior performance in both sample means and IQR. The reason for the classification of B97-D3/aDZ as inferior to B3LYP-D3/aDZ and M05-2X/aDZ, despite a seemingly excellent sample mean for ∆COM signed errors, is the larger spread of the errors for B97-D3/aDZ; indeed, the range in the ∆COM signed error is 0.32 ˚ A for B9714

ACS Paragon Plus Environment

Page 14 of 34

Page 15 of 34

D3/aDZ versus 0.16 ˚ A for B3LYP-D3/aDZ and 0.11 ˚ A for M05-2X/aDZ. While this wider distribution of errors cancels fortuitously for B97-D3/aDZ to produce a very low MSE (0.00 ˚ A), the mean absolute error (MAE) for this model chemistry is nearly double that of B3LYPD3/aDZ, with MAE = 0.06, 0.03 ˚ A, respectively. Despite the presence of some cases with errors slightly larger than the target value for B97-D3/aDZ, each density functional is able to produce equilibrium geometries of the desired accuracy level for a significant percentage of the A21 complexes when paired with the aDZ basis set.

3.2

Prediction of Optimal Intermolecular Separation in NBC7x and HBC6 Interaction Energy Scans Interaction Energy (kcal/mol)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

0

B3LYP-D3 B97-D3 M05-2X CCSD(T)/CBS

–5

–16 –18 –20

–10

R

–22

–15

–24 4.0

–20

4.1

–25 3.4

3.6

3.8

4.0

4.2

4.4

4.6

4.8

Intermolecular Separation, R (Å)

Figure 4: Scans of the non-counterpoise–corrected interaction energy (unCP IE) along the radial separation coordinate R in the formamidine dimer (HBC6-3; inset shown) computed with B3LYP-D3 (red), B97-D3 (blue) and M05-2X (green) using the cc-pVDZ basis set. The interpolated optimal intermolecular separation for each curve is indicated with a vertical dotted line in the same colors. For reference, a curve constructed from the CCSD(T)/CBS benchmark IEs at each value of R is presented in black. As illustrated in Fig. 4 for the formamidine dimer, optimal intermolecular separations (Req ) were interpolated from radial potential scans constructed for the 13 complexes in the HBC6 and NBC7x test sets using both counterpoise-corrected (CP) and uncorrected (unCP) 15

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation

(a) NBC7x: CP

(b) NBC7x: unCP

(c) HBC6: CP

(d) HBC6: unCP

0.2

∆COM Signed Error (Å)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 34

0.1 0.0 –0.1

B3LYP-D3

B97-D3

M05-2X

0.10 0.05 0.00

–0.05 –0.10

DZ

TZ

aDZ

aTZ

DZ

TZ

aDZ

aTZ

Figure 5: Box-and-whisker plots depicting the distribution of signed error in interpolated optimal center-of-mass displacement (∆COM) for radial interaction energy curves in the NBC7x (a & b) and HBC6 (c & d) test sets. For both test sets, box-and-whisker plots representing curves constructed from both counterpoise-corrected (CP; left panels a & c) and uncorrected (unCP right panels, b & d) IEs are given. Whiskers encompass the full range of ∆COM signed errors for the indicated test set, correction scheme, and model chemistry, and boxes illustrate the first (Q1), second (median, black bar), and third quartiles (Q3) of these data; additionally, the mean signed error for each data set is indicated with a black square. For reference, a dotted line indicates 0.0 ˚ A, and three levels of shading are provided: light grey encompassing ±0.1 ˚ A, medium grey encompassing ±0.05 ˚ A, and medium-dark ˚ grey encompassing ±0.01 A.

16

ACS Paragon Plus Environment

Page 17 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

interaction energies, computed using each combination of density functional and basis set examined above. The CCSD(T)/CBS reference curve, from which the reference Req value is interpolated, is also shown; for a complete set of equivalent figures (108 total), please refer to the Supporting Information. Provided in Fig. 5 are box-and-whisker plots describing the distribution of signed errors of these interpolated minima for each DFT model chemistry, as compared to the minima interpolated from reference curves. Regardless of the choice of BSSE treatment (either CP or unCP), interpolated minima for curves in the NBC7x test set exhibit slightly larger signed errors than those for HBC6 curves; while all model chemistries produce MSE ≤ ±0.06 ˚ A for both CP and unCP curves within the HBC6 test set (and twelve model chemistries with MSE ≤ ±0.01 ˚ A!), several model chemistries for NBC7x complexes yield MSE slightly outside this range, albeit still within the target of Req ≤ ±0.1 ˚ A. Errors in interpolated Req for CP-curves are largely independent of basis set size or augmentation for both NBC7x and HBC6, with the lone exeption of HBC6 curves with TZ exhibiting improved MSE and IQR over DZ. For unCP curves, however, the quality of interpolated Req is sensitive to both ζ-level and augmentation. Considering next the performance of individual model chemistries, B3LYP-D3 and B97D3 perform about the same (CP curves) or better (unCP curves) for NBC7x complexes with a triple-ζ basis set, regardless of augmentation, while for M05-2X, the double-ζ basis sets are better. For HBC6, interpolated minima are quite good for all model chemistries considered, although errors are slightly larger for unCP curves using the DZ basis set. For both test sets, B3LYP-D3 and B97-D3 exhibit excellent performance for Req ; among the 32 total combinations of these two functionals, the four examined basis sets, and two possible BSSE treatments, MSE ≤ 0.01 ˚ A for Req of 16 model chemistries, 0.01 ≤ MSE ≤ 0.05 ˚ A for Req of 11 model chemistries, and the remaining five model chemistries all produce interpolated Req values with 0.05 ≤ MSE ≤ 0.1 ˚ A. While M05-2X yields high quality interpolated Req for HBC6 systems, this functional tends on average to underbind for CP curves of NBC7x systems, as well as for unCP curves with triple-ζ basis sets (see, e.g., Figs. S-76–S-82 and S-

17

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

111–117 in the Supplemental Information), leading to a slight overestimation of the optimal intermolecular separation distance Req ; the MSE for these cases ranges between 0.07–0.1 ˚ A. Interestingly, NBC7x systems with π − π stacking (NBC7x-1 & 7; sandwich benzene and pyridine dimers, respectively) seem particularly susceptible to this drastic under-binding by M05-2X, exhibiting deviations from reference interaction energies as large as +2 kcal mol−1 in the neighborhood of the minima of CP curves, a full factor of four larger than deviations exhibited by either B97-D3 or B3LYP-D3 (see, e.g., Figs. S-62 and S-67 in the Supplemental Information). The T-shaped counterparts to these complexes (NBC7x-2 & 8 for benzene and pyridine dimers, respectively), while slightly less underbound, still exhibit significant deviations from CCSD(T)/CBS IEs of nearly 1 kcal mol−1 in the neighborhood of the curve minima (Figs. S-63 and S-68). While these T-shaped complexes are much more realistically described by M05-2X with double-ζ basis sets when not employing the counterpoise-correction procedure, the sandwich configurations are similarly underbound for unCP curves even with double-ζ, and both sandwich and T-shaped configurations are severerly underbound in unCP curves with triple-ζ basis sets. Strangely, this underbinding of systems involving π-stacking or CH−π interactions by M05-2X is not present in other dispersion-dominated NBC7x systems, e.g., methane dimer, where M05-2X underbinds either by only tenths of kcal mol−1 for CP curves (e.g., Fig. S-66) or nearly perfectly reproduces CCSD(T)/CBS IEs for unCP curves (e.g., Fig. S-73); these results are consistent with values reported previously in Ref. 25. Finally, we examine in particular the performance of aDZ model chemistries for unCP curves, as our optimizations of A21 complexes did not employ BSSE corrections, and aDZ was found to perform similarly to aTZ. For these larger systems, MSE ≤ 0.02 ˚ A for HBC6 curves, and MSE ≤ 0.05 ˚ A for NBC7x curves constructed with B3LYP-D3 and M05-2X. B97-D3/aDZ performs only slightly worse for NBC7x curves, with MSE = -0.08 ˚ A; each of these model chemistries, however, produce Req MSEs within the target range of accuracy. This indicates that the high quality of DFT/aDZ geometries observed in Section 3.1 for

18

ACS Paragon Plus Environment

Page 18 of 34

Page 19 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

the optimization of A21 systems is likely to generalize to the optimization of systems as large as those in HBC6 or NBC7x, and perhaps somewhat larger. We must note, however, that in systems that are much larger, long-range effects neglected by M05-2X and manybody dispersion interactions neglected by all of the approaches tested here may become significant. 33–37

4

Conclusions

We have shown that each of B3LYP-D3, B97-D3, and M05-2X density functionals paired with Dunning’s aug-cc-pVDZ (aDZ) basis set combine accuracy and reasonable computational expense for producing equilibrium geometries of 21 small bimolecular van der Waals complexes from the A24 test set. Each DFT/aug-cc-pVDZ level of theory performs well compared to CCSD(T)/CBS references: both B3LYP-D3/aDZ and M05-2X/aDZ consistently yield equilibrium geometries with very small least root-mean-square displacement (LRMSD) and center-of-mass displacement signed error (∆COM SE), each within 0.05 ˚ A on average. B97-D3/aDZ nearly as good, but exhibits slightly larger range of ∆COM SE. We have also shown that these DFT/aDZ combinations are capable of reproducing optimal intermolecular separation distances (Req ) interpolated from radial interaction energy scans of 13 larger complexes in the HBC6 and NBC7x test sets. Minima interpolated from curves in both HBC6 and NBC7x test sets constructed from non–counterpoise-corrected (unCP) interaction energies computed using DFT/aDZ are of similarly high quality, with B3LYP-D3/aDZ and M05-2X/aDZ yielding minima within 0.05 ˚ A of CCSD(T)/CBS, while B97-D3/aDZ is again nearly as good but slightly less reliable for unCP curves in NBC7x. Overall the analysis of optimized A21 systems, together with the quality of interpolated NBC7x and HBC6 minima, indicate that each of these DFT/aDZ combinations are well suited to produce equilibrium geometries of given conformations of bimolecular van der Waals complexes of diverse binding motif.

19

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

In the course of this work, we developed a software tool to maximally align the approximate geometries optimized with DFT against the CCSD(T) benchmark geometries, to obtain the LRMSD metric for each A21 complex analyzed above. This tool, consisting of Python implementations of two general algorithms solving the maximal alignment problem, together with all data presented here and all Python code necessary to perform the above data analysis and visualization, are available free of charge via an open-source GitHub repository at www.github.com/cdsgroup/dftoptbench-si. All software contained in this repository can be executed without local installation via the Jupyter Hub cloud server, or cloned to be used offline.

5

Acknowledgements

The authors gratefully acknowledge financial support from Bristol-Myers Squibb, and from the U.S. National Science Foundation through grant CHE-1566192. A.A. was supported jointly by the National Science Foundation and the NASA Astrobiology Program, under the NSF Center for Chemical Evolution, CHE-1504217.

6

Associated Content

Additional computational details, including a discussion of the A21 benchmark geometries, assessment of optimization quality, and a discussion of our solution to the point set registration problem; tables detailing reference data for A21, HBC6, and NBC7x test sets; tables and figures detailing optimization quality for each model chemistry examined; interaction energy scans for HBC6 and NBC7x systems, computed with each model chemistry; summary figures detailing the quality of interpolated potential energy minima for both reference and density functional potential energy scans of HBC6 and NBC7x systems. Additionally, Cartesian geometries of A21 systems optimized with each model chemistry examined here are provided as a zipped archive of XYZ files. All of these materials are available free of charge 20

ACS Paragon Plus Environment

Page 20 of 34

Page 21 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

via the Internet at http://pubs.acs.org. Finally, all data generated in this work is available at http://www.github.com/cdsgroup/dftoptbench-si, organized as a Python database using the Pandas and Jupyter Notebook software.

21

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

References (1) Khaliullin, R. Z.; Head-Gordon, M.; Bell, A. T. An Efficient Self-Consistent Field Method for Large Systems of Weakly Interacting Components. J. Chem. Phys. 2006, 124, 204105. (2) Khaliullin, R. Z.; Cobar, E. A.; Lochan, R. C.; Bell, A. T.; Head-Gordon, M. Unravelling the Origin of Intermolecular Interactions Using Absolutely Localized Molecular Orbitals. J. Phys. Chem. A 2007, 111, 8753–8765. (3) Khaliullin, R. Z.; Bell, A. T.; Head-Gordon, M. Analysis of Charge Transfer Effects in Molecular Complexes Based On Absolutely Localized Molecular Orbitals. J. Chem. Phys. 2008, 128, 184112. (4) Horn, P. R.; Sundstrom, E. J.; Baker, T. A.; Head-Gordon, M. Unrestricted Absolutely Localized Molecular Orbitals for Energy Decomposition Analysis: Theory and Applications to Intermolecular Interactions Involving Radicals. J. Chem. Phys. 2013, 138, 134119. (5) Jeziorski, B.; Moszynski, R.; Szalewicz, K. Perturbation Theory Approach to Intermolecular Potential Energy Surfaces of van der Waals Complexes. Chem. Rev. 1994, 94, 1887–1930. (6) Szalewicz, K. Symmetry-adapted Perturbation Theory of Intermolecular Forces. WIREs Comput. Mol. Sci. 2012, 2, 254–272. (7) Hohenstein, E. G.; Sherrill, C. D. Wavefunction Methods for Noncovalent Interactions. WIREs Comput. Mol. Sci. 2012, 2, 304–326. (8) Sherrill, C. D. Energy Component Analysis of π Interactions. Acc. Chem. Res. 2013, 46, 1020–1028.

22

ACS Paragon Plus Environment

Page 22 of 34

Page 23 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

(9) Parrish, R. M.; Sherrill, C. D. Spatial Assignment of Symmetry Adapted Perturbation Theory Interaction Energy Components: The Atomic SAPT Partition. J. Chem. Phys. 2014, 141, 044115. (10) Parrish, R. M.; Parker, T. M.; Sherrill, C. D. Chemical Assignment of SymmetryAdapted Perturbation Theory Interaction Energy Components: The Functional-Group SAPT Partition. J. Chem. Theory Comput. 2014, 10, 4417–4431. (11) Parrish, R. M.; Sitkoff, D. F.; Cheney, D. L.; Sherrill, C. D. The Surprising Importance of Peptide Bond Contacts in Drug-Protein Interactions. Chem. Eur. J. 2017, 23, 7887– 7890. (12) Bakr, B. W.; Sherrill, C. D. Analysis of Transition State Stabilization by Non-Covalent Interactions in the Houk-List Model of Organocatalyzed Intermolecular Aldol Additions Using Functional-Group Symmetry-Adapted Perturbation Theory. Phys. Chem. Chem. Phys. 2016, 18, 10297–10308. (13) Smith, D. G. A.; Burns, L. A.; Patkowski, K.; Sherrill, C. D. Revised Damping Parameters for the D3 Dispersion Correction to Density Functional Theory. J. Phys. Chem. Lett. 2016, 7, 2197–2203. (14) Sherrill, C. D.; Takatani, T.; Hohenstein, E. G. An Assessment of Theoretical Methods for Nonbonded Interactions: Comparison to Complete Basis Set Limit Coupled-Cluster Potential Energy Curves for the Benzene Dimer, the Methane Dimer, Benzene-Methane, and Benzene-H2 S. J. Phys. Chem. A 2009, 113, 10146–10159. ˇ y, J.; Hobza, P.; Salahub, D. R. Density Functional Theory Aug(15) Jureˇcka, P.; Cern´ mented with an Empirical Dispersion Term. Interaction Energies and Geometries Of 80 Noncovalent Complexes Compared with Ab Initio Quantum Mechanics Calculations. J. Comput. Chem. 2007, 28, 555–569.

23

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

(16) Vald´es, H.; Klus´ak, V.; Pitoˇ n´ak, M.; Exner, O.; Star´ y, I.; Hobza, P.; Rul´ıˇsek, L. Evaluation of the Intramolecular Basis Set Superposition Error in the Calculations of Larger Molecules: [n]Helicenes and Phe-Gly-Phe Tripeptide. J. Comput. Chem. 2008, 29, 861–870. ˇ aˇc, J. Evaluation of Composite Schemes for CCSDT(Q) (17) Demovicova, L.; Hobza, P.; Rez´ Calculations of Interaction Energies of Noncovalent Complexes. Phys. Chem. Chem. Phys. 2014, 16, 19115–19121. ˇ aˇc, J.; Simova, L.; Hobza, P. CCSD[T] Describes Noncovalent Interactions Better (18) Rez´ Than the CCSD(T), CCSD(TQ), and CCSDT Methods. J. Chem. Theory Comput. 2013, 9, 364–369. ˇ ac, J.; Riley, K. E.; Hobza, P. Benchmark Calculations of Noncovalent Interactions (19) Rez´ of Halogenated Molecules. J. Chem. Theory Comput. 2012, 8, 4285. ˇ aˇc, J.; Riley, K. E.; Hobza, P. S66: A Well-Balanced Database of Benchmark Inter(20) Rez´ action Energies Relevant to Biomolecular Structures. J. Chem. Theory Comput. 2011, 7, 2427–2438. ˇ ˇ y, J.; Hobza, P. Benchmark Database of Accurate (MP2 (21) Jureˇcka, P.; Sponer, J.; Cern´ and CCSD(T) Complete Basis Set Limit) Interaction Energies of Small Model Complexes, DNA Base Pairs, and Amino Acid Pairs. Phys. Chem. Chem. Phys. 2006, 8, 1985–1993. (22) Thanthiriwatte, K. S.; Hohenstein, E. G.; Burns, L. A.; Sherrill, C. D. Assessment of the Performance of DFT and DFT-D Methods for Describing Distance Dependence of Hydrogen-Bonded Interactions. J. Chem. Theory Comput. 2011, 7, 88–96. (23) Marshall, M. S.; Burns, L. A.; Sherrill, C. D. Basis Set Convergence of the CoupledCCSD(T )

cluster Correction, δM P 2

: Best Practices for Benchmarking Non-covalent Interac-

24

ACS Paragon Plus Environment

Page 24 of 34

Page 25 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

tions and the Attendant Revision of the S22, NBC10, HBC6, and HSG Databases. J. Chem. Phys. 2011, 135, 194102. (24) Raghavachari, K.; Trucks, G. W.; Pople, J. A.; Head-Gordon, M. A 5th-Order Perturbation Comparison of Electron Correlation Theories. Chem. Phys. Lett. 1989, 157, 479–483. ´ Sumpter, B. G.; Sherrill, C. D. Density(25) Burns, L. A.; V´azquez-Mayagoitia, A.; Functional Approaches to Noncovalent Interactions: A Comparison of Dispersion Corrections (DFT-D), Exchange-Hole Dipole Moment (XDM) Theory, and Specialized Functionals. J. Chem. Phys. 2011, 134, 084107. (26) Burns, L. A.; Marshall, M. S.; Sherrill, C. D. Appointing Silver and Bronze Standards for Noncovalent Interactions: A Comparison of Spin-Component-Scaled (SCS), Explicitly Correlated (F12), and Specialized Wavefunction Approaches. J. Chem. Phys. 2014, 141, 234111. (27) Grimme, S. Semiempirical GGA-Type Density Functional Constructed With a LongRange Dispersion Correction. J. Comput. Chem. 2006, 27, 1787–1799. (28) Becke, A. D. Density-Functional Thermochemistry. III. The Role of Exact Exchange. J. Chem. Phys. 1993, 98, 5648–5652. (29) Stephens, P. J.; Devlin, F. J.; Chabalowski, C. F.; Frisch, M. J. Ab Initio Calculation of Vibrational Absorption And Circular Dichroism Spectra Using Density Functional Force Fields. J. Phys. Chem. 1994, 98, 11623–11627. (30) Zhao, Y.; Schultz, N. E.; Truhlar, D. G. Design of Density Functionals by Combining the Method Of Constraint Satisfaction with Parametrization For Thermochemistry, Thermochemical Kinetics, and Noncovalent Interactions. J. Chem. Theory Comput. 2006, 2, 364–382.

25

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

(31) Grimme, S.; Antony, J.; Ehrlich, S.; Krieg, H. A Consistent and Accurate Ab Initio Parametrization of Density Functional Dispersion Correction (DFT-D) for the 94 Elements H-Pu. J. Chem. Phys. 2010, 132, 154104. ˇ aˇc, J.; Hobza, P. Describing Noncovalent Interactions Beyond the Common Ap(32) Rez´ proximations: How Accurate Is the Gold Standard, CCSD(T) at the Complete Basis Set Limit? J. Chem. Theory Comput. 2013, 9, 2151–2155. (33) Ambrosetti, A.; Ferri, N.; DiStasio Jr., R. A.; Tkatchenko, A. Wavelike charge density fluctuations and van der Waals interactions at the nanoscale. Science 2016, 351, 1171– 1176. (34) Ambrosetti, A.; Reilly, A. M.; DiStasio Jr., R. A.; Tkatchenko, A. Long-range correlation energy calculated from coupled atomic response functions. J. Chem. Phys. 2014, 140, 18A508. (35) Ambrosetti, A.; Alf`e, D.; DiStasio Jr., R. A.; Tkatchenko, A. Hard Numbers for Large Molecules: Toward Exact Energetics for Supramolecular Systems. J. Phys. Chem. Lett. 2014, 5, 849–855. (36) Misquitta, A. J.; Spencer, J.; Stone, A. J.; Alavi, A. Dispersion interactions between semiconducting wires. Phys. Rev. B. 2010, 82, 075312. (37) Dobson, J. F.; White, A.; Rubio, A. Asymptotics of the Dispersion Interaction: Analytic Benchmarks for van der Waals Energy Functionals. Phys. Rev. Lett. 2006, 96, 073201. (38) Becke, A. D.; Johnson, E. R. Exchange-hole Dipole Moment and the Dispersion Interaction. J. Chem. Phys. 2005, 122, 154104. (39) Johnson, E. R.; Becke, A. D. A Post-Hartree-Fock Model of Intermolecular Interactions: Inclusion of Higher-order Corrections. J. Chem. Phys. 2006, 124, 174104.

26

ACS Paragon Plus Environment

Page 26 of 34

Page 27 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

(40) Kong, J.; Gan, Z. T.; Proynov, E.; Freindorf, M.; Furlani, T. R. Efficient Computation of the Dispersion Interaction with Density-functional Theory. Phys. Rev. A 2009, 79, 042510. (41) Dion, M.; Rydberg, H.; Schr¨oder, E.; Langreth, D. C.; Lundqvist, B. I. van der Waals Density Functional for General Geometries. Phys. Rev. Lett. 2004, 92, 246401. (42) Langreth, D. C.; Dion, M.; Rydberg, H.; Schroder, E.; Hyldgaard, P.; Lundqvist, B. I. van der Waals Density Functional Theory with Applications. Int. J. Quantum Chem. 2005, 101, 599–610. ´ Kong, L.; Lundqvist, B. I.; Langreth, D. C. Higher-accuracy van (43) Lee, K.; Murray, E.; der Waals Density Functional. Phys. Rev. B 2010, 82, 081101(R). (44) Vydrov, O. A.; Voorhis, T. V. Nonlocal van der Waals Density Functional: The Simpler the Better. J. Chem. Phys. 2010, 133, 244103. (45) Zhao, Y.; Truhlar, D. G. Density Functionals for Noncovalent Interaction Energies Of Biological Importance. J. Chem. Theory Comput. 2007, 3, 289–300. (46) Hohenstein, E. G.; Chill, S. T.; Sherrill, C. D. Assessment of the Performance of the M05-2X and M06-2X Exchange-Correlation Functionals for Noncovalent Interactions in Biomolecules. J. Chem. Theory Comput. 2008, 4, 1996–2000. ˇ aˇc, J.; Pulay, P.; Hobza, P. Accuracy of Quan(47) Sedlak, R.; Janowski, T.; Pitoˇ nak, M.; Rez´ tum Chemical Methods for Large Noncovalent Complexes. J. Chem. Theory Comput. 2013, 9, 3364–3374. (48) Perdew, J. P.; Schmidt, K. Jacob’s ladder of density functional approximations for the exchange-correlation energy. AIP Conference Proceedings 2001, 577, 1. (49) Grimme, S.; Hansen, A.; Brandenburg, J. G.; Bannwarth, C. Dispersion-Corrected Mean-Field Electronic Structure Methods. Chem. Rev. 2016, 116, 5105–5154. 27

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

(50) Parrish, R. M.; Burns, L. A.; Smith, D. G. A.; Simmonett, A. C.; DePrince, A. E.; Hohenstein, E. G.; Bozkaya, U.; Sokolov, A. Y.; Di Remigio, R.; Richard, R. M.; Gonthier, J. F.; James, A. M.; McAlexander, H. R.; Kumar, A.; Saitow, M.; Wang, X.; Pritchard, B. P.; Verma, P.; Schaefer, H. F.; Patkowski, K.; King, R. A.; Valeev, E. F.; Evangelista, F. A.; Turney, J. M.; Crawford, T. D.; Sherrill, C. D. Psi4 1.1: An OpenSource Electronic Structure Program Emphasizing Automation, Advanced Libraries, and Interoperability. J. Chem. Theory Comput. 2017, 13, 3185–3197. (51) Shao, Y.; Gan, Z.; Epifanovsky, E.; Gilbert, A. T. B.; Wormit, M.; Kussmann, J.; Lange, A. W.; Behn, A.; Deng, J.; Feng, X.; Ghosh, D.; Goldey, M.; Horn, P. R.; Jacobson, L. D.; Kaliman, I.; Khaliullin, R. Z.; Kus, T.; Landau, A.; Liu, J.; Proynov, E. I.; Rhee, Y. M.; Richard, R. M.; Rohrdanz, M. A.; Steele, R. P.; Sundstrom, E. J.; Woodcock, H. L.; Zimmerman, P. M.; Zuev, D.; Albrecht, B.; Alguire, E.; Austin, B.; Beran, G. J. O.; Bernard, Y. A.; Berquist, E.; Brandhorst, K.; Bravaya, K. B.; Brown, S. T.; Casanova, D.; Chang, C.; Chen, Y.; Chien, S. H.; Closser, K. D.; Crittenden, D. L.; Diedenhofen, M.; DiStasio, R. A.; Do, H.; Dutoi, A. D.; Edgar, R. G.; Fatehi, S.; Fusti-Molnar, L.; Ghysels, A.; Golubeva-Zadorozhnaya, A.; Gomes, J.; Hanson-Heine, M. W. D.; Harbach, P. H. P.; Hauser, A. W.; Hohenstein, E. G.; Holden, Z. C.; Jagau, T.; Ji, H.; Kaduk, B.; Khistyaev, K.; Kim, J.; Kim, J.; King, R. A.; Klunzinger, P.; Kosenkov, D.; Kowalczyk, T.; Krauter, C. M.; Lao, K. U.; Laurent, A. D.; Lawler, K. V.; Levchenko, S. V.; Lin, C. Y.; Liu, F.; Livshits, E.; Lochan, R. C.; Luenser, A.; Manohar, P.; Manzer, S. F.; Mao, S.; Mardirossian, N.; Marenich, A. V.; Maurer, S. A.; Mayhall, N. J.; Neuscamman, E.; Oana, C. M.; Olivares-Amaya, R.; O’Neill, D. P.; Parkhill, J. A.; Perrine, T. M.; Peverati, R.; Prociuk, A.; Rehn, D. R.; Rosta, E.; Russ, N. J.; Sharada, S. M.; Sharma, S.; Small, D. W.; Sodt, A.; Stein, T.; Stueck, D.; Su, Y.; Thom, A. J. W.; Tsuchimochi, T.; Vanovschi, V.; Vogt, L.; Vydrov, O.; Wang, T.; Watson, M. A.; Wenzel, J.; White, A.; Williams, C. F.; Yang, J.; Yeganeh, S.; Yost, S. R.; You, Z.; Zhang, I. Y.; Zhang, X.; 28

ACS Paragon Plus Environment

Page 28 of 34

Page 29 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

Zhao, Y.; Brooks, B. R.; Chan, G. K. L.; Chipman, D. M.; Cramer, C. J.; Goddard, W. A.; Gordon, M. S.; Hehre, W. J.; Klamt, A.; Schaefer, H. F.; Schmidt, M. W.; Sherrill, C. D.; Truhlar, D. G.; Warshel, A.; Xu, X.; Aspuru-Guzik, A.; Baer, R.; Bell, A. T.; Besley, N. A.; Chai, J.; Dreuw, A.; Dunietz, B. D.; Furlani, T. R.; Gwaltney, S. R.; Hsu, C.; Jung, Y.; Kong, J.; Lambrecht, D. S.; Liang, W.; Ochsenfeld, C.; Rassolov, V. A.; Slipchenko, L. V.; Subotnik, J. E.; Voorhis, T. V.; Herbert, J. M.; Krylov, A. I.; Gill, P. M. W.; Head-Gordon, M. Advances in Molecular Quantum Chemistry Contained in the Q-Chem 4 Program Package. Mol. Phys. 2015, 113, 184–215. (52) Johnson, E. R.; Becke, A. D.; Sherrill, C. D.; DiLabio, G. A. Oscillations in Metageneralized-gradient Approximation Potential Energy Surfaces for Dispersion-bound Complexes. J. Chem. Phys. 2009, 131, 034111. (53) Whitten, J. L. Coulombic Potential-Energy Integrals and Approximations. J. Chem. Phys. 1973, 58, 4496–4501. (54) Dunlap, B. I.; Connolly, J. W. D.; Sabin, J. R. Applicability of LCAO-X-alpha Methods to Molecules Containing Transition-metal Atoms - Nickel Atom and Nickel Hydride. Int. J. Quantum Chem. Symp. 1977, 11, 81. (55) Dunlap, B. I.; Connolly, J. W. D.; Sabin, J. R. On Some Approximations in Applications of Xα Theory. J. Chem. Phys. 1979, 71, 3396–3402. (56) Feyereisen, M.; Fitzgerald, G.; Komornicki, A. Use of Approximate Integrals in Ab Initio Theory. An Application in MP2 Calculations. Chem. Phys. Lett. 1993, 208, 359–363. (57) Vahtras, O.; Alml¨of, J.; Feyereisen, M. W. Integral Approximations for LCAO-SCF Calculations. Chem. Phys. Lett. 1993, 213, 514–518. (58) Rendell, A. P.; Lee, T. J. Coupled-cluster Theory Employing Approximate Integrals:

29

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

An Approach to Avoid the Input/Output and Storage Bottlenecks. J. Chem. Phys. 1994, 101, 400–408. (59) Kendall, R. A.; Fruchtl, H. A. The Impact of the Resolution of the Identity Approximate Integral Method On Modern Ab Initio Algorithm Development. Theor. Chem. Acc. 1997, 97, 158–163. (60) Weigend, F. A Fully Direct RI-HF Algorithm: Implementation, Optimized Auxiliary Basis Sets, Demonstration of Accuracy and Efficiency. Phys. Chem. Chem. Phys. 2002, 4, 4285–4291. (61) Dunning, T. H. Gaussian basis sets for use in correlated molecular calculations. I. The atoms boron through neon and hydrogen. J. Chem. Phys. 1989, 90, 1007–1023. ˇ aˇc, J.; Hobza, P. Describing Noncovalent Interactions beyond the Common Approx(62) Rez´ imations: How Accurate Is the “Gold Standard,” CCSD(T) at the Complete Basis Set Limit? J. Chem. Theory Comput. 2013, 9, 2151–2155. (63) East, A. L. L.; Allen, W. D. The Heat of Formation of NCO. J. Chem. Phys. 1993, 99, 4638–4650. (64) Cs´asz´ar, A. G.; Allen, W. D.; Schaefer, H. F. In Pursuit of the Ab Initio Limit for Conformational Energy Prototypes. J. Chem. Phys. 1998, 108, 9751–9764. (65) Halkier, A.; Helgaker, T.; Jørgensen, P.; Klopper, W.; Koch, H.; Olsen, J.; Wilson, A. K. Basis-set Convergence in Correlated Calculations On Ne, N2 , and H2 O. Chem. Phys. Lett. 1998, 286, 243–252. (66) Koch, H.; Fern´andez, B.; Christiansen, O. The Benzene-Argon Complex: A Ground and Excited State Ab Initio Study. J. Chem. Phys. 1998, 108, 2784–2790. (67) Tsuzuki, S.; Honda, K.; Uchimaru, T.; Mikami, M.; Tanabe, K. Origin of Attraction

30

ACS Paragon Plus Environment

Page 30 of 34

Page 31 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

and Directionality of the π-π Interaction: Model Chemistry Calculations of Benzene Dimer Interaction. J. Am. Chem. Soc. 2002, 124, 104–112. (68) Sinnokrot, M. O.; Valeev, E. F.; Sherrill, C. D. Estimates of the Ab Initio Limit for π-π Interactions: The Benzene Dimer. J. Am. Chem. Soc. 2002, 124, 10887–10893. (69) Jureˇcka, P.; Hobza, P. On the Convergence of the (∆E CCSD(T) - ∆E MP2 ) Term for Complexes with Multiple H-bonds. Chem. Phys. Lett. 2002, 365, 89–94. ˇ (70) Hobza, P.; Sponer, J. Toward True DNA Base-Stacking Energies: MP2, CCSD(T), And Complete Basis Set Calculations. J. Am. Chem. Soc. 2002, 124, 11802–11808. (71) Sinnokrot, M. O.; Sherrill, C. D. High-Accuracy Quantum Mechanical Studies of π-π Interactions in Benzene Dimers. J. Phys. Chem. A 2006, 110, 10656–10668. (72) Boese, A. D.; Martin, J. M. L.; Klopper, W. Basis Set Limit Coupled Cluster Study of H-bonded Systems and Assessment of More Approximate Methods. J. Phys. Chem. A 2007, 111, 11122–11133. (73) Janowski, T.; Pulay, P. High Accuracy Benchmark Calculations On the Benzene Dimer Potential Energy Surface. Chem. Phys. Lett. 2007, 447, 27–32. (74) Pitoˇ na´k, M.; Janowski, T.; Neogr´ady, P.; Pulay, P.; Hobza, P. Convergence of the CCSD(T) Correction Term for the Stacked Complex Methyl Adenine-Methyl Thymine: Comparison with Lower-Cost Alternatives. J. Chem. Theory Comput. 2009, 5, 1761– 1766. (75) Takatani, T.; Hohenstein, E. G.; Malagoli, M.; Marshall, M. S.; Sherrill, C. D. Basis Set Consistent Revision of the S22 Test Set of Noncovalent Interaction Energies. J. Chem. Phys. 2010, 132, 144104. (76) Faver, J. C.; Benson, M. L.; He, X.; Roberts, B. P.; Wang, B.; Marshall, M. S.; Kennedy, M. R.; Sherrill, C. D.; Merz Jr., K. M. Formal Estimation of Errors in Com31

ACS Paragon Plus Environment

Journal of Chemical Theory and Computation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

puted Absolute Interaction Energies of Protein-Ligand Complexes. J. Chem. Theory Comput. 2011, 7, 790–797. (77) Carrell, E. J.; Thorne, C. M.; Tschumper, G. S. Basis Set Dependence of Higher-order Correlation Effects in π-type Interactions. J. Chem. Phys. 2012, 136, 014103. (78) Burns, L. A.; Marshall, M. S.; Sherrill, C. D. Comparing Counterpoise-Corrected, Uncorrected, and Averaged Binding Energies for Benchmarking Noncovalent Interactions. J. Chem. Theory Comput. 2014, 10, 49–57. (79) Singh, N. J.; Min, S. K.; Kim, D. Y.; Kim, K. S. Comprehensive Energy Analysis for Various Types of π-Interaction. J. Chem. Theory Comput. 2009, 5, 515–529. (80) Boys, S. F.; Bernardi, F. The Calculation of Small Molecular Interactions by the Differences of Separate Total Energies. Some Procedures with Reduced Errors. Mol. Phys. 1970, 19, 553–566. (81) van der Walt, S.; Colbert, S. C.; Varoquaux, G. The NumPy Array: A Structure for Efficient Numerical Computation. Computing in Science and Engineering 2011, 13, 22–30.

32

ACS Paragon Plus Environment

Page 32 of 34

Page 33 of 34 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Theory and Computation

Quality of Optimized Supramolecular Geometry 0.1 0.0 –0.1 B97-D3/aDZ